Depth gauge

The depth gauge is the interactive explainer that shows how far a single reply falls down the cascade before a rung is certain enough to act.

The metaphor

The ladder is drawn as a water column. The surface is expensive: rung 7, full LLM reasoning. The floor is classical and free: rung 0, header checks. A reply enters at the bottom and climbs only as far as it needs; every rung it passes is silent and costs nothing.

The waterline is the confidence gate (0.55 at rung 3). A rung below the waterline resolves the reply and the rungs above it never wake. A reply that no rung below the waterline is certain about keeps climbing — similar examples (4), the trained classifier (5), the meaning-match judge (6) — until it either resolves or surfaces to the LLM.

What the gauge shows

• The cascade column — seven rungs, lit down to the one that resolved the reply; the readout names the rung, the intent, the confidence and the gate.
• The probe — paste a reply and watch it classified live through the running cascade.
• The thread panel — replays a demo conversation turn by turn, with the drafted reply, the rung that produced each action, and the pursuit state.
• Capabilities and learning — which rungs are active and what the learning loop has collected so far.

On this site

The landing page reproduces the gauge’s hero column with pre-computed traces from the demo thread. The live gauge runs the same cascade answering in real time.

Rung placements and their numbers are documented in the cost ladder. The experiments behind them are in research & experiments.